ePubWU Institutional RepositoryMax Roser and Jesus Crespo Cuaresma Why is Income Inequality Increasing in the Developed World? Article (Accepted for Publication) (Refereed) Original

ePubWU Institutional Repository

Max Roser and Jesus Crespo Cuaresma

Why is Income Inequality Increasing in the Developed World?

Article (Accepted for Publication)(Refereed)

Original Citation:

Roser, Max and Crespo Cuaresma, Jesus ORCID: https://orcid.org/0000-0003-3244-6560

(2016)

Why is Income Inequality Increasing in the Developed World?

Review of Income and Wealth, 62 (1).

pp. 1-27. ISSN 00346586

This version is available at: https://epub.wu.ac.at/4529/Available in ePubWU: May 2015

ePubWU, the institutional repository of the WU Vienna University of Economics and Business, isprovided by the University Library and the IT-Services. The aim is to enable open access to thescholarly output of the WU.

This document is the version accepted for publication and — in case of peer review — incorporatesreferee comments. There are minor differences between this and the publisher version which couldhowever affect a citation.

http://epub.wu.ac.at/

https://orcid.org/0000-0003-3244-6560

https://epub.wu.ac.at/4529/

http://epub.wu.ac.at/

Why is income inequality increasingin the developed world?∗

Max Roser† Jesus Crespo Cuaresma‡

Abstract

We address empirically the factors affecting the dynamics of income inequality amongindustrialized economies. Using a panel for 32 developed countries spanning the lastfour decades, our results indicate that the predictions of the Stolper-Samuelson theoremconcerning the effects of international trade on income inequality find support in the dataif we concentrate on imports from developing countries as a trade measure, as theorywould imply. We find that democratization, the interaction of technology and educationand changes in the relative power of labour unions affect inequality dynamics robustly.

Keywords: Inequality, international trade, FDI, globalization, institutions

JEL codes: D63, I24, F10

∗The authors would like to thank Tony Atkinson, Conchita D’Ambrosio (the editor) and two anonymousreferees for helpful comments to earlier versions of this paper. Max Roser would like to thank the Institute forNew Economic Thinking at Oxford for their hospitality during the preparation of this study and the Office ofthe Vice Rector for Research at the University of Innsbruck for financial support.†Institute for New Economic Thinking, Oxford Martin School, University of Oxford. Address: Eagle House,

Walton Well Road, Oxford. Email: [email protected]‡Department of Economics, Vienna University of Economics and Business (WU); Wittgenstein Centre for

Demography and Global Human Capital (WIC); World Population Program, International Institute of AppliedSystems Analysis (IIASA) and Austrian Institute of Economic Research (WIFO). Address: Welthandelsplatz1, 1020 Vienna (Austria). Email: [email protected].

1 Introduction

In this paper we analyse empirically the factors behind the dynamics of inequality in indus-trialized countries during the last four decades. Although we focus on the interplay betweenglobalization and income inequality, a subject where theory has provided clear-cut predictionsbut empirical work often delivers contradictory results, we aim at a more comprehensive as-sessment of the drivers of inequality in advanced economies than those offered by the empiricalstudies hitherto.

The available empirical literature tends to concentrate individually on particular theories andmechanisms, while abstracting from assessing simultaneously other arguably important drivingforces. Given the fact that the theoretical literature on the determinants of inequality offersa manifold of explanatory frameworks which are not necessarily mutually exclusive, most ofthe contributions in the literature lack important controls and thus tend to offer fragmentaryevidence. The strong differences in the results of the existing empirical studies can be thustraced back not only to differences in the coverage of countries and time periods and in theeconometric methods used, but also to the partial nature of many of the models used to assessthe determinants of inequality.

Our analysis improves on existing studies in several ways. Firstly, we use the largest sample ofdata available for industrialized economies, which spans data for 32 countries over four decadesand employ modern econometric methods based on dynamic panel data models which explicitlyaccount for potential endogeneity problems (see Blundell and Bond, 1998). These statistical is-sues have not been taken into account by many of the existing studies on the subject, which mayhave led to sizeable biases in parameter estimates. On the other hand, we concentrate on mod-els which comprise different explanatory factors for inequality dynamics, instead of includingexplanatory variables which exclusively relate to single theoretical mechanisms. In particular,we include explanatory factors related to theories of technological change, international tradeand political conditions and institutions as determinants of income inequality.

In our analysis we put particular effort in assessing the role played by international tradein shaping inequality dynamics. In order to mirror the theoretical setting implied by theHeckscher-Ohlin model - and especially by its corollary, the Stolper-Samuelson theorem - weassess explicitly the effect of different trade flows on inequality measures instead of concentrat-ing on aggregate trade openness measures. In particular, we construct measures of trade flowsbetween developing countries and developed economies, thus remaining closer to the theoreticalframework of the Heckscher-Ohlin model, which focuses on countries with different factor en-dowments. Our results robustly show that, in line with the predictions of the Stolper-Samuelsontheorem and in contrast to many existing empirical studies, imports from low-income countriesdo increase inequality in industrialized economies. In addition, the size of the government,political regime changes, unionization and the interplay between education expansion and tech-nological change appear as further important determinants of within-country variation in theGini index.

The paper is organized as follows. Section 2 presents a brief survey of the existing empiricalstudies dealing with the determinants of income inequality. Section 3 shows descriptive evidenceon the dynamics of inequality in our sample of 32 industrialized economies. Section 4 presentsthe theoretical explanations provided by the literature on the determinants of inequality and

1

describes the variables used to approximate the determinants implied by such theories. Section5 sets up the econometric model and presents the data, while the results of the estimation ofour econometric models are presented in section 6. Section 7 concludes.

2 The Determinants of Income Inequality: A Bird’s Eye

Review of the Recent Literature

Unveiling the determinants of inequality empirically has been the focus of a large numberof studies in the last decades. Several recent publications empirically study the role playedby different socioeconomic variables as determinants of inequality both within and betweencountries, concentrating often on single explanatory mechanisms.

Some of these studies explicitly focus on the impact of international trade and globalizationon income inequality. Reuveny and Li (2003) assesses the effects of trade and democracy oninequality. Using data for a broad panel of countries, they find that both trade and democracydecrease income inequality. Dreher and Gaston (2008) assess whether globalization increasedinequality using an index which merges several measures related to various aspects of globaliza-tion. Mahler (2004), on the other hand, examines the causes of income inequality and finds onlyweak evidence concerning the effect of forces related to international integration. On the otherhand, variables that relate to domestic influences – the partisan balance of national cabinets,electoral turnout, union density, and the centralization of wage-setting institutions – are foundnot to have a strong effect on inequality. Jaumotte et al. (2008) investigate the rise of incomeinequality in 20 industrialized countries and 31 developing countries and also focus on the ef-fects of globalization. They find that trade globalization has no effect on inequality measuresbut financial globalization does increase inequality. Increases in foreign direct investment (FDI)are thus found to be associated with rising inequality.

Another group of studies focuses on the distributional effects of political and institutionalfactors. Calderon et al. (2005) use a large panel of 121 countries spanning the three decadesbetween 1970 and 2000 and highlight the relevance of labour market regulations. Baccaro (2011)finds that neither trade unionism nor collective bargaining tend to be statistically associatedwith lower income equality. The differences in the nature of the political interference of statesin the market and its effect on inequality is at the heart of the analysis carried out by Ruedaand Pontusson (2000), in the spirit of the so-called Varieties of Capitalism paradigm (see Halland Soskice, 2001, as the core text of this school). The results of Rueda and Pontusson (2000)concerning the determinants of the ratio of the ninth to the first income deciles for 16 OECDcountries indicate that union density is the most important factor behind inequality dynamicsand that “[its] effects are consistently egalitarian and they are greater than those of any otherindependent variable” (Rueda and Pontusson, 2000, p. 352). Pontusson et al. (2002) focus onthe role of partisanship and labour market institutions as determinants of inequality in differentparts of the income distribution. They find that unionization, centralization of wage bargainingand public-sector employment primarily affect the distribution of wages by affecting the relativeposition of unskilled workers, while the upper half of the distribution is primarily influenced bythe partisanship of the government.

2

Oliver (2008) studies the dynamics of wage inequality in a panel of 14 OECD member statesbetween 1980 and 2002 and finds that industry-wide wage scales are systematically related towithin-country inequality changes. Beramendi and Cusack (2008) use a panel of 13 OECDcountries and focus on the persistence in inequality measures. They explain these long-termdifferences as a consequence of the work of unions and political actors. Koeniger et al. (2007)study the causes of income inequality in 11 OECD member states concentrating on labourmarket variables and conclude that labor market institutions can account for a large part ofthe change of income inequality between countries.

This compact review of some of the most relevant empirical studies in the recent literaturereveals a multitude of partly contradictory results based on different samples, methods and(fragmentary) theoretical frameworks. We aim at providing a more comprehensive view of thedriving factors of inequality than the studies hitherto, basing our analysis on adequate methodsto assess the potential endogeneity problems of the econometric specification and combininginformation of the different frameworks available in theory.

3 Income Inequality in the Developed World: Data and

Dynamics

Before specifying econometric models, we start by exploring the overall long-run dynamics ofinequality in advanced economies. Data on economic inequality is notoriously problematic.One of the major pitfalls of existing cross-country datasets is the combination of sources whichare based on different definitions of income inequality. Data may differ in terms of the includedsources of income (capital versus labour income) or of recipients (households versus workers).Datasets are thus difficult to compare and data that combine different sources are often severelycriticized.

- Include Figure 1 here -

The dataset by Deininger and Squire (1996) combined data for many countries and for sometime was the most widely used dataset and a number of the studies surveyed in the previoussection are based on this source. The downside of the dataset by Deininger and Squire (1996)is that Gini indices differ in terms of income definitions (e.g. gross or net) and reference units(e.g., households or persons). After the critique from Atkinson and Brandolini (2001) otherdatasets (or corrections of the Deininger-Squire data) tend to be preferred in empirical studiescomprising information for several countries.


Variability of the inequality measure in the time dimension appears particularly important todisentangle the effect of its determinants and to assess the potential endogeneity problems inthe relationships under scrutiny. In this respect, the Estimated Household Income Inequality(EHII) dataset (Galbraith and Kum, 2005), which corrects the Deininger-Squire data by uti-lizing the data on income dispersion sourced from the United Nations Industrial Development

3

Organization (UNIDO), appears as the natural choice for our purpose, taking into accountthe deficiencies of other existing datasets. The EHII dataset interpolates missing data gaps inthe original sources using the data sourced from UNIDO. This is achieved using regressions ofincome inequality (Gini) indices from the Deininger-Squire dataset on the UNIDO variable andother covariates, including controls for the type of data source.1 Thereby, a dataset is assembledfor (partly estimated) household gross income Gini indices in a comprehensive and consistentmanner. The dataset is a large unbalanced panel for 154 countries from 1963 through 2002with nearly 3,200 country-year observations.

In spite of its broad coverage and homogeneous approach to interpolating missing data, somelimitations of the EHII data should be noted. To the extent that the figures in the EHIIdata are partly estimated, they may be subject to biases. By concentrating on informationon manufacturing wages in order to achieve a consistent interpolation exercise, the measurescovered by the EHII data may be inaccurate in countries where the inequality characteristicsof the industrial sector are not representative for the whole income distribution. Althoughthe EHII dataset has been shown to capture trends in inequality in a very reliable manner,it is known to partly fail at capturing volatility in capital income for top percentiles of theincome distribution (see e.g. Galbraith et al., 2014). These caveats should be kept in mindwhen interpreting the results of our estimation.

Our analysis concentrates on a panel of developed industrialized countries, for which relativelylong time series of inequality data are available. All member states of the OECD and all coun-tries that are classified as an advanced economy by the IMF2 for which data are available areincluded in the analysis. The group of 32 countries is composed by Australia, Austria, Bel-gium, Canada, Chile, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary,Iceland, Ireland, Israel, Italy, Japan, South Korea, Luxembourg, Mexico, Netherlands, NewZealand, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden, Turkey, UnitedKingdom and the United States.3

The overall dynamics of inequality for the full sample of industrialized countries used in ouranalysis are depicted in Figure 1 presenting the average Gini index for the 32 countries. Figure1 shows both the unweighted average and the population-weighted average of Gini indices.Figure 2, on the other hand, shows the values of the inequality index by country.4

The average level of inequality in developed countries has risen since the beginning of theeighties, while the population-weighted average indicates that such an increase started in themid-sixties if we take into account that relatively large countries increased their Gini indexalready prior to the birth of this global trend. The variation of the Gini index between countriesexceeds that within countries, although the statistics in Table A2 indicate that stark differencesare present in the country-specific dynamics of income inequality. While decreases in the Giniindex took place in Iceland and Belgium, all other countries experienced changes in the direction

1See the technical appendix for a brief description of the interpolation method.2The list can be found at: http://www.imf.org/external/pubs/ft/weo/2009/01/weodata/groups.htm.3From the original 39 countries which fulfilled the selection criterion (OECD member state or advanced

economy according to the IMF), Hong Kong, Taiwan, Malta, Cyprus, Singapore, Switzerland and Estonia hadto be excluded due to missing data. This reduction leaves 32 countries in the panel for which in some cases wehave data ranging back to the sixties.

4Country-specific descriptive statistics of the inequality measure for the full time period considered are shownin Table A1 (Appendix). Table A2 (Appendix) presents evidence concerning the change in the Gini index sincethe eighties.

4

of a more unequal distribution of income across individuals. The most sizeable increases in theGini index took place in countries in Central and Eastern Europe, and double-digit percentageincreases in inequality since the eighties is more the rule than the exception among industrializedeconomies.

4 The Drivers of Income Inequality

In order to specify our empirical model in terms of the choice of explanatory variables, we needto assess the nature of the main drivers of inequality from a theoretical point of view. Thereis a wide range of theories that pinpoint the causes of income inequality. The central economictheories that study the driving forces of inequality dynamics tend to relate to one of threedifferent broad explanatory factors: (a) technological change, that affects inequality through itseffect on the skill premium; (b) international trade, which changes relative product prices andthus the wage distribution and (c) the consequences of changes of the political institutions andpolitical conditions.

4.1 Technological Change

The seminal paper by Kuznets (1955) analyses the dynamics of inequality during the transitionfrom an agricultural to an industrialized economy. The argument following this view pointsout that inequality rose when the transition to industrialization set in. The industrializationprocess leads to a large demand for high-skilled entrepreneurs and engineers, who were in shortsupply at this moment in history. At a later stage of industrialization in the first half of thetwentieth century, inequality declined rapidly. Kuznets (1955) explains this phenomenon as aconsequence of the general increase in the level of education, which led to a decline of the skillpremium and at the same time boosted the productivity of workers. Advances in technologythus increase the demand for high-skilled labour on the one hand and on the other hand provideincentives for investing in education. Jan Tinbergen referred to this phenomenon as the racebetween education and technology.

Another reason for the rising skill-premium in industrialized countries is related to the decliningdemand for low-skilled labour. This development takes place as many industrialized countriesundergo a transition from industry-based economies to service economies, which itself is partlya consequence of globalized international trade. The declining demand for low-skilled work inmanufacturing is at least partly a consequence of rising imports of labour intensive goods fromlow-income countries.5 The influence of factor-biased technological change on factor demandand factor prices is discussed in the framework of labour market models by Krugman (2000),Acemoglu (2002) and Acemoglu (2003).

4.2 International Trade and Foreign Direct Investment

International trade and technological change as root causes for income inequality dynamics areintertwined in a way that makes it difficult to distinguish exactly their differential effects. First

5See also Feenstra (1998, pages 41-42).

5

of all, international trade in its modern form is just possible because of modern technology.Especially developments in transport and communication are prerequisites for modern interna-tional trade. Secondly, international trade changes the demand for high-skilled and low-skilledworkers in different economies and thereby also alters the skill premium. This is precisely theprediction by standard trade theory following the reasoning of the Heckscher-Ohlin model. Itstates that a country exports those products that use factors of production which are availablein abundance in the economy. This means that labour-abundant developing countries exportmostly products that are labour-intensive in production and import products which are skill-and capital-intensive in production. The outsourcing of labour-intensive production to low-wagecountries therefore causes a lower demand for low-skilled workers and a higher demand for high-skilled workers in industrialized economies and rises the skill premium. Location matters thusin the sense that production sites are allocated in a way as to profit from these comparativeadvantages. Production requiring low skill levels is therefore shifted to low-wage countries. Inthe framework of the Heckscher-Ohlin model, changes in wages are the consequence of changesin product prices. International trade extends markets, which implies that prices of productsthat are exported are higher than they would be if that product was not sold in internationalmarkets. Product prices of imported goods, in turn, are lower than the national price of thesame products would be, which is the reason for trading them.

As international trade has grown extensively – in real terms it has multiplied by 45 between1949 and 2008 – there is much discussion of adverse effects of this development on the distri-bution of income. This is a direct prediction of standard trade theory. The Stolper-Samuelsontheorem – an immediate corollary of the Heckscher-Ohlin model – predicts that the onset orthe intensification of international trade changes factor remunerations. In the framework of amodel in which high-skilled labour and low-skilled labour are considered to be two differentfactors of production this means that the distribution of incomes between these two factors isaltered by the intensification of international trade. The reasoning behind this prediction isstraightforward: as the prices of imported products fall, the wages in the import-competingsector also fall. In contrast, the rising prices of those goods that are exported cause higherwages in the exporting sectors of the economy. Interestingly, the Stolper-Samuelson theoremstates that the translation of product prices to factor prices is not proportional – the result-ing dispersion of wages is even greater than the dispersion of the related product prices,6 animplication known as the the magnification effect.

The trend of increasing globalization has other effects on the wage structure which go beyondthe mechanisms proposed by the Heckscher-Ohlin paradigm. It possibly narrows the room forbargaining and therefore lowers the possible impact of wage bargaining institutions, since theyface more elastic labour demand curves, particularly for the low-skilled workers.

4.3 Political Conditions, Institutions and the Labour Market

Besides international trade and technological progress, a variety of labour market characteristicshave been considered important to explain the dynamics of income distribution. Theoriststhat incorporate such political conditions and institutions have an understanding of the labourmarket that goes beyond the interplay of market forces. The level and the dynamics of income

6See Samuelson (1949).

6

inequality are thus not only assumed to be determined by market interactions but the marketitself is shaped by political agency. In this indirect way, political forces possibly matter for thedistribution of incomes. While the political forces that possibly shape labour market structuresinclude a wide variety of institutional characteristics of the labour market and the politicalsphere, the study of these effects is considerably limited by the availability of appropriate data.

Since the objective of trade unions is to raise the relative wages of their members, unionizationdynamics may affect wage dispersion within the economy positively or negatively, dependingon what union members would be paid if they were not unionized. Blau and Kahn (1996) notethat it is particularly the low-paid jobs that are addressed by unions. In many countries7 unionstend to raise wages for the low-paid jobs just as a national minimum wage would. Freeman(1982) shows that wage dispersion within the group of unionized workers is smaller than in thegroup of non-unionized workers. In this sense, the equalization of the pay of their membersis an unambiguous income-equalizing effect of unions. In addition to these direct equalizingeffects there is an indirect effect of unions that affects non-unionized workers. Employers ofnon-unionized workers try to avert the threat of unionization by raising the wage of theseworkers to the union level. This anticipated wage adoption is referred to as the “threat effect”and is well documented, as the review in Western and Rosenfeld (2011) shows. In addition,while the existence of unions is expected to be related to a more equal distribution of wages,its effect on the distribution of household income is ambiguous. In particular, if unions reducelabour demand and increase non-employment, a higher share of jobless households would leadto raising income inequality. The results in Checchi and Garcia-Penalosa (2008) and Checchiand Garcia-Penalosa (2010) point indeed in this direction.

5 Empirical Specification: Explaining Income Inequality

Dynamics

5.1 The Basic Model

The basic econometric specification on which we base our inference concerning the drivers ofinequality is given by

INEQit = φINEQi,t−1 + β′xit + µi + λt + νit, (1)

where INEQit is the measure of income inequality of country i in period t and xit is a k-dimensional vector of explanatory variables, that may be strictly exogenous, predetermined orendogenous covariates. These variables are linked to the inequality variable through the pa-rameters which are collected in the column vector β. Our model specification assumes country-specific fixed effects (µi) and global shocks, summarized by period-specific effects (λt). Theremaining shock, vit, is assumed to be an i.i.d. distributed error term with constant variance.

The dynamic structure of the panel data model in (1) implies that there is a built-in correlationbetween the lagged endogenous variable and the error term that lead to biases if the estimates ofthe parameters in the model are obtained using least-square methods. Starting with Anderson

7Austria, Germany, Italy, Sweden, and Switzerland are documented in the study by Blau and Kahn (1996).

7

and Hsiao (1981) and Anderson and Hsiao (1982), methods based on instrumental variableshave been developed to overcome the statistical problems involved in the estimation of (1). Thepopular dynamic panel data estimator put forward by Arellano and Bond (1991) proposes to usesuitably lagged values of yt as instruments in the first-differenced specification in the frameworkof generalized method of moments (GMM) estimation. The Arellano-Bond estimator is howevera suboptimal choice if the variable being explained is highly persistent, since lagged levels ofa highly persistent variable are weak instruments for its first difference. Blundell and Bond(1998) propose to consider both the corresponding equation in levels and its first-differencedcounterpart to formulate an extended set of overidentifying restrictions for the GMM estimator.The estimator proposed by Blundell and Bond (1998) has become standard when dealing withdynamic panel models of highly persistent variables.

The econometric study is conducted for the aforementioned panel of 32 developed industrializedcountries. Panel unit root tests for the dependent variable were undertaken before engagingin the econometric analysis of the causes of inequality. The Im-Pesaran-Shin test (Im et al.,2003) yields a Zt−bar-value of -3.85, with a corresponding p-value of 0.0001.8 This implies thatwe consider our panel stationary and, accordingly, we estimate models where the inequalitymeasure enters in levels instead of in first differences.

5.2 From Theory to Data

As discussed above, we use Gini index data from the EHII dataset as the dependent variable inour estimated specifications. In order to approximate technological change, we use data fromHeston et al. (2011) to calculate total factor productivity (TFP) for all countries in our samplebased on a standard Cobb-Douglas production function, Yt = AtF (Kt, Lt) = AtK

αt L

1−αt . We

use labour force data (Lt) and the standard value of α of 0.3 to retrieve estimates of TFP (At)after obtaining estimates of the capital stock (Kt) using the perpetual inventory method.9

Trade volume estimates are routinely used as explanatory variables in inequality models inorder to quantify the effect of international trade.10 Many studies use data on total tradeopenness (see for example Baccaro, 2011; Reuveny and Li, 2003; Lee and Slotsve, 2001), whichare readily available in many datasets. Such data are however probably misleading for thequestion at hand. The bulk of international trade nowadays is not trade between nationswith different factor endowments but intra-industry trade between industrialized countries thatfeature similar factor endowments. Intra-industrial trade can have very different causes andconsequences as compared to inter-industry trade, and can be explained by models that donot account for factor endowments.11 Therefore, the part of international trade which is in

8This is the result for the Dickey–Fuller regression specified without a time trend. Including a time trendincreases the Zt−bar-value to -7.50, for which the corresponding p-value is essentially zero.

9As a robustness check, we also reestimated our models using total factor productivity data from the Confer-ence Board dataset. This source provides TFP data for a broad of countries, albeit only ranging back to 1990.The most important results of our analysis are not strongly affected by the use of these alternative estimates oftotal factor productivity. The results of this robustness test are available from the authors upon request.

10For a discussion, see Krugman (2000). Opposing the claim that just prices matter, Krugman (2000) arguesthat trade volumes represent changes in demand and supply that cause product price changes and just thesechanges reflect the relation between international trade and income inequality.

11The theoretical reasons for intra-industry trade are increasing returns and the search for an enlarged market.See Helpman and Krugman (1987) for a basic reference on the causes and consequences of intra-industry trade.

8

fact intra-industry trade is not necessarily related to rising income inequality in the way it ispredicted by the Stolper-Samuelson theorem. To determine the effects of international trade onincome inequality in industrialized countries, we have to utilize data on the volume of importswhose production is low-skilled labour intensive and exports whose production is high-skilledlabour intensive. As such, these data are not readily available. Rueda and Pontusson (2000),for instance, propose to use data on imports from least developed countries (LDCs), which isactually available from the OECD, albeit only starting in 1980. It can be argued, however,that grouping all developing countries as the relevant aggregate may not be precise enough.Several of the LDCs included are primarily exporters of raw materials (e.g. Saudi Arabia, theIslamic Republic of Iran, Venezuela or Equatorial Guinea). As trade in raw materials is notdue to different factor-endowments, it is not theoretically related to the dynamics predicted bythe Stolper-Samuelson theorem.

In order to overcome these shortcomings, we calculate the relevant international trade variableusing the Bilateral Trade (v2.01) Dataset of the Correlates of War (COW) Project. This datasetcontains information on bilateral trade volumes for a vast panel of countries from 1870 to 2006.We concentrate on trade between the developed countries in our sample and those LDCs thatare not major oil exporters or whose exports are not predominately raw materials and mineralresources. The data on export composition which is necessary for this exercise are obtainedfrom Isham et al. (2005), who classify countries by their export structure and identify thosecountries that predominantly export raw materials. Four major oil producing countries whichare not studied in Isham et al. (2005) (Russia, Kuwait, Libya and the United Arab Emirates)are added to the list. The reduced list of LDCs that are not predominantly exporters of crudematerials contains 101 countries. An observation of the relevant trade variable for a given yearand a given industrialized country is the sum of imports from all of these 101 countries to thecorresponding developed economy, normalized by the GDP of the importing economy. Thissample of countries is denoted as ’developing countries, no oil’ throughout this study.

The Stolper-Samuelson theorem does not only consider the impact of imports but also the effectsof exports. Therefore we also calculated the exports of our sample of developed countriesto all developing countries. Figure 3 gives an overview of the magnitude of imports fromdeveloping countries. As an FDI measure, we employ data on inward and outward FDI stocksas a percentage of GDP that are sourced from the United Nation’s World Investment Report.


Since the institutional setting – and in particular institutions related to the labour market – hasbeen proposed as an important determinant of inequality, we use variables which summarizesuch information in our empirical specification. Union density data are used to proxy forthe power of unions. Data on union density for 34 countries ranging from 1960 to 2007 areavailable from the Database on Institutional Characteristics of Trade Unions, Wage Setting,State Intervention and Social Pacts - (ICTWSS). As the dataset has some missing values, wecomplemented it with data sourced from the OECD.

Autocratic regimes have the possibility to concentrate wealth and income in their governmentcircles. It can be argued that democratic regimes are more likely to provide for most members ofthe society. We control for such political institutions using the measure of democracy (polity2)from the Polity-IV dataset. The degree of interference of the state into the economy is measured

9

making use of the total tax share of GDP or, alternatively, the share of public expenditures onGDP. Unemployment data from Armingeon et al. (2010) and Armingeon and Romana (2007)are used, as well as data on female labour participation which are sourced from the OECD’sLabour Force Statistics (MEI).

In addition to the theoretical forces put forward in Section 3, other determinants of the wagestructure need to be controlled for when assessing the driving forces of income inequality dy-namics. Inflation possibly increases inequality, as it induces divergence between persistent andfast-adapting wages. The data on inflation were obtained from the World Bank’s World Devel-opment Indicators dataset. We also control for the effects that the average level of income ina country has on inequality using PPP-adjusted income per capita sourced from Heston et al.(2011). We further control for business cycle dynamics by including GDP growth as a covariatein the specification – the data are again from Heston et al. (2011).

In order to assess the effects of immigration on income distribution dynamics, we control fornet migration – the number of immigrants minus the number of emigrants – as a share of totalpopulation. Data originating from the OECD and Eurostat are taken from Samanni et al.(2010).12 An important control variable is the level of education of the population. Data onmean years of schooling for the population aged 15 and above from the Barro-Lee database areused for this purpose. In addition, data on population growth and the dependency ratio aresourced from the World Development Indicators.

6 Estimation Results

6.1 Inequality and International Trade

We start by investigating the link between international trade and inequality. Table 1 presentsseven estimated models which differ solely on the international trade variable that is included inthe model (imports from developing economies excluding oil exporters, imports from developingeconomies, total imports, exports to developing economies, total exports and total trade, allnormalized by GDP).

In addition to the corresponding trade variable and the lag of the dependent variable, our modelsinclude three measures to capture the macroeconomic conditions of the economy: the GDP percapita, its square and GDP per capita growth. Two further control variables are includedin the baseline specification: the total tax share of GDP and the mean years of schoolingof the population over 15. All explanatory variables are lagged by one year. The models areestimated using system-GMM methods. GDP per capita, its square and GDP per capita growthare considered endogenous and are instrumented by one further lag in the framework of theGMM estimation of the model.13

- Include Table 1 here -

12If data from the OECD and Eurostat were available, we calculate the average. If none of these two sourcesare available, data are taken from the World Development Indicators.

13We decided to keep a relatively small set of instruments in our estimation. Enlarging the set improves theresults of the Sargan test but does not affect our conclusions significantly. Results are available from the authorsupon request.

10

Columns 1 to 6 in Table 1 show the estimation results for the baseline model extended by our sixdifferent trade measures. Of all of the trade variables that are positively and significantly relatedto income inequality the impact of imports from non-oil-exporting LDCs is the largest amongall trade covariates. Our results in Table 1 indicate that, as predicted by the Heckscher-Ohlinmodel and the Stolper-Samuelson theorem, trade with developing countries in fact worsensthe income distribution (that is, increases the Gini index) in developed countries. Importsfrom all LDCs have a smaller and less significant impact on the wage distribution, with acoefficient which is just significant at the 10% level and an impact on inequality correspondingto a 6% increase in the Gini index relative to its standard deviation if LDC imports rise byone standard deviation. For the same relative change in imports from non-oil-exporting LDCsinequality increases by roughly 25% of its standard deviation. The exclusion of raw materialexporting countries appears thus crucial and unveils a sizeable effect of imports on inequalityas predicted by theory.

Overall trade openness appears highly significant but has a quantitatively much smaller influ-ence on inequality. This highly aggregated measure seems to obscure the theoretical workingof the impact of trade on inequality and only raises inequality by 8% of its standard deviationwhen it rises by one standard deviation. In column 7 we present the estimates of the modelwith trade openness as an explanatory variable using fixed-effects OLS, which also confirm theresults of the GMM estimation concerning the effect of this variable.

None of the covariates relating to the economic development of the economy have a robustimpact on the dynamics of income inequality. The parameter estimates for the linear andsquared terms of GDP per capita, included to check for the nonlinear effect predicted by Kuznets(1955), do not confirm an inverse U-shaped relationship between income and inequality.14


Table 2 shows the results for alternative specifications of the model including the variable whichmeasures imports from non-oil-exporting LDCs. In columns 1 to 3 we present the results ofsmaller models, while the model in column 4 corresponds to the specification in column 1 oftable 1. In column 5 we present the results of a model where we substitute the tax share ofGDP for the share of public expenditures on GDP. The effect of a change in the government sizecovariate is qualitatively similar across these two variables and the change in the specificationdoes not affect the rest of the results strongly.15 The results presented in column 6 correspondto a model extended by including (the lagged level of) union density and inflation, as wellas FDI variables. Inflation and union density have significant effects that correspond to themechanisms expected from a theoretical perspective. More unionized labour markets tend tobe related to less inequality, while higher inflation tends to increase the Gini index. In spiteof the fact that we lose an important part of our sample by using these two covariates in the

14It should be noted that in many specifications the AR(2) test gives evidence of second-order autocorrelationin the residuals of the first-differenced model. This result is driven by the observations at the beginning of thesample, as was confirmed by reestimating the models for subsamples. The main results of the analysis remainunchanged if the sample is restricted to the last two decades, where the AR(2) test is not significant. Thismakes us confident that the linkages unveiled in our empirical analysis are indeed robust and not just driven bypotentially misspecified regression models.

15Given the larger coverage of the tax share data, we stick to this measure of government size for the rest ofthe specifications.

11

specification, their inclusion does not qualitatively change our conclusions concerning the effectsof the trade variable. We lose more than half of our sample when we include FDI variablesin the model. The size and significance of the effect of our trade variable in this specificationare reduced, but it remains significant at the 10% level. A marginally significant effect of FDIinward stocks implying that more FDI leads to more inequality is also present in the data.

Our clear-cut results concerning the effect of trade on inequality contrast with the estimates ofother studies which assess the same relationship. Lundberg and Squire (2003), Baccaro (2011)and Figini and Gorg (2011) do not find a strong influence of trade on inequality employingtrade openness as an independent variable. Reuveny and Li (2003) also employ overall tradeopenness as a regressor and find a significantly income-equalizing effect of trade for their globalsample as well as for two subsamples of developing and OECD countries, although the effect isquantitatively rather small (if trade openness rises by one standard deviation, the Gini index isreduced by 14.22% of its standard deviation). Jaumotte et al. (2008) find that an increase in theexport-to-GDP ratio reduces inequality while the share of imports on GDP is insignificant. Bynarrowing their sample and only including developed economies, they find that a rising shareof imports from developing countries reduces inequality. Mahler (2004) finds that imports fromLDCs and other measures of globalization are not significantly related to wage inequality. Thisfinding is similar to that of Rueda and Pontusson (2000) and Oliver (2008), who both findthat the degree of openness with less developed countries is not significantly associated to thedynamics of wage inequality in OECD countries. Beramendi and Cusack (2008) study the effectof LDC imports on wage inequality in OECD countries and find the opposite effect as theorywould predict. Dreher and Gaston (2008) focus on the question on whether globalization hasincreased inequality and employ the KOF-index as a measure of globalization. They find that arise of this index is significantly associated with a rise of inequality. The effects of globalizationon within-country inequality are also analyzed by Bussmann et al. (2005). Their primary focusis the influence of FDI and they do not find a link between its dynamics and inequality. Inaddition to this, the coefficients of both of their trade related measures are not significantlydifferent from zero. For these reasons, they conclude that globalization has in fact not adverselyaffected national income inequality.

The fact that international trade for developed countries is to a large extent intra-industrialtrade explains such results and the difference with our estimates using only trade with non-oil-exporting LDCs. The same applies to the pooling of all developing countries as tradingpartners without making an explicit difference between exporters of raw materials and countrieswhich are specialized in low-wage manufactured goods. The study by Pontusson et al. (2002)gets closer to our definition of the trade variable. They use data on imports from LDCs andexclude from this panel the OPEC member states. Their results, however, do not confirm asignificant positive relationship between inequality and trade. The use of different time periodshas also strong effects on the nature of the association found. Reestimating our model for twosubperiods (pre-1990 and post-1990) results on significant and strong effects only for the morerecent subsample. 16

16It should be noted that the poor results on identifying the relationship implied by the Stolper-Samuelsontheorem in previous studies might also be due to inappropriate estimation strategies. Many studies do not takeadvantage of up-to-date dynamic panel data estimators and studies which do so either do not include tradeflows as a independent variable (as Calderon et al., 2005) or use data with an a priori relatively low informationsignal (as Dreher and Gaston, 2008), who employs an aggregated index of globalization).

12

The study by Meschi and Vivarelli (2009) can be seen as complementary to our study since itis methodologically and theoretically closely related but in contrast to our study concentrateson the dynamics of inequality in developing countries. Meschi and Vivarelli (2009) presentevidence based on 65 developing countries over the period spanning the eighties and ninetiesand find that imports from and exports to developed countries significantly worsen the incomedistribution in middle-income countries.


Table 3 show the estimation results from models that include further control variables. Theinclusion of the dependency ratio, population growth and immigration does not alter the resultsqualitatively and the variables themselves do not affect inequality in a statistically significantmanner. We furthermore find an insignificant effect of technological progress as measured bythe Solow residual. In Table 4, we provide the estimation results of models where we moveaway from pure linear effects of technology and education by including also an interaction termbetween these variables. Such an interaction can be thought of as modelling Tinberger’s “racebetween education and technology”. The results indicate that the negative effect of educationon the inequality measure is reduced in economies with relatively high total factor productivity,thus lending support that the joint dynamics of productivity and skilled labour are able to partlyexplain changes in the Gini index over the sample studied. The sign of the estimated parametersshow that in relatively underdeveloped economies (in terms of total factor productivity), theincome-equalizing effects of education are stronger than in highly developed countries.


In further model specifications which were estimated but are not shown, we find that changes ofthe unemployment rate are not associated with the dynamics of inequality. This confirms theresults in other studies (Oliver (2008), for instance) and can be explained by two countervailingeffects that cancel each other out. On the one hand rising unemployment puts higher pressureon low income workers as their bargaining power diminishes and their wages are kept low, thusincreasing inequality. On the other hand, rising unemployment is likely to imply that formerlylow wage workers become unemployed. In this sense, increases in unemployment could beassociated with a reduction of inequality among wage earners.17

6.2 Inequality and Democratization

Table 5 on page 27 shows in its first column the results of a model that includes the level ofdemocracy as an extra regressor in our specification, using the polity2 variable from the Polity-IV dataset. This variable ranges from -10 (authoritarian regime) to +10 (democratic regime)and was transformed to the range [0,10]. The estimated negative coefficient for the variable in-dicates that democratization goes along with decreasing inequality. Dreher and Gaston (2008),on the contrary, find that democratization is related to a rise in income inequality. They justifythese results arguing that an increase in democracy is associated with market-oriented reforms,which increase inequality.

17These results are available from the authors upon request.

13


These contradictory results regarding the relationship between democracy and income inequal-ity can be better understood by assessing empirically the effects of changes in the political regimefor different groups of countries. We classified all the countries into three different groups. Allformer members of the Warsaw Pact and their successor states were grouped together (Hungary,Poland, the Slovak Republic and the Czech Republic), a second group comprises all countriesfrom Western Europe (Iceland, Austria, Belgium, Denmark, Finland, France, Germany, Greece,Ireland, Italy, Slovenia, Luxembourg, Malta, Netherlands, Norway, Portugal, Spain, Sweden,Switzerland and United Kingdom) and the remaining countries are subsumed to a third group.18

For the Western European group, four countries were not always considered to be fully demo-cratic in the sense of the polity2 index: Portugal, Spain, France and Greece. France wasnever rated 10 after 1958 primarily because of weak “executive constraints” (see http://www.

systemicpeace.org/polity/France2008.pdf). The other three countries underwent drasticdemocratization processes over our observation period. Portugal was rated -9 in 1973 and afterthe Carnation Revolution in 1974 and the end of the authoritarian dictatorship of the EstadoNovo under Antonio de Oliveira Salazar, the political regime was rated with 9 as of 1976. Acomparable change of political regime happened in neighboring Spain. Before the death ofFrancisco Franco, Spain was rated -7 (1973). Three years later, Spain was rated 9 – even beforethe proclamation of the constitution in 1978. Greece’s transition from the dictatorship underGeorgios Papadopoulos to a democracy is reflected in the change from -7 in 1973 to a rating of8 in 1975. The democratization processes that took place in these countries went hand in handwith a trend towards a more equal distribution of income, as can be seen in Figure 4, whichdepicts the average Gini index for the three groups described above. Notice that, on the otherhand, the strong democratization processes which took place in Eastern Europe happened inparallel to an increase in the Gini index.


We estimate a model with different parameters attached to the democracy index for each oneof these group, which is presented in Table 5. Since there is a possible endogeneity problembecause, as Muller (1995) shows, falling levels of inequality make democratization more likely,we specifying the democracy variable as endogenous in the GMM framework in addition tousing its lagged value (as with all other covariates). The coefficient for the group of ex-Sovietcountries is negative and resembles the results that Dreher and Gaston (2008) find for theirentire sample. Our result for this subsample can be interpreted as reflecting the transition froma socialist economy to a market economy, that was accompanied with a rise of the inequalityof incomes. This argument is in line with Krugman et al. (1999). Although our result isinsignificant for non-European countries, changes in the level of democracy are found to beassociated with a decline of inequality for the rest of the countries in our sample.

We also estimated regressions based on specifications that include a political risk indicator.In particular, we use the measures proposed by Kuncic (2014) which are available for the

18Although Slovenia was not part of the Warsaw Pact, its inclusion in this group could be argued given thepolitical history of the country. The results presented are not influenced by the exclusion of Slovenia of theWarsaw Pact group, as robustness checks that we conducted clearly show. These are available from the authorupon request.

14

period 1990-2003. The results, not shown here but available upon request, show that onlyimprovements in “economic quality”, an index which is related to the well-functioning of marketsand property rights enforcement, appear to be associated with rising inequality.

6.3 Inequality and the Size of the State

The tax share of GDP is a significant determinant of income inequality throughout all specifi-cations. Increases in the variable tend to be robustly associated with decreases of inequality.Several channels explain such a result. A rising share of tax revenue is likely to be associatedwith an increase in the number of public sector employees and this in turn is likely to contributeto income equalization as the income inequality in the public sector is smaller than in the pri-vate sector. This precise relationship – an equalizing effect of higher public sector employment– is found by Calderon et al. (2005) and Pontusson et al. (2002). On the other hand, a risingshare of taxes possibly implies a declining incentive to aim for higher incomes as those are in-creasingly taxed. The tax share of GDP tends to be strongly correlated with regulation of thelabour market such as minimum-wage laws19 and stronger labour protection. Unemploymentbenefits are also likely to be higher in states with a larger tax share of GDP. Koeniger et al.(2007) find that more generous unemployment benefits in fact lower income inequality. Thiscan be explained by the higher reservation wage that forces employers to adjust their low wagesaccordingly. Simple correlation analysis across countries confirms such associations as potentialmechanisms driving the effect of government size on inequality.


Rueda and Pontusson (2000) analyze the dynamics of inequality separately for what theycall Liberal Market Economies (LME) and Social Market Economies (SME). They classifyeconomies as belonging to one or the other category by considering several indicators, includingthe labour decommodification index. We classify the countries in our sample as belonging toone or the other category relying on their mean tax share of GDP. By considering all countrieswith a higher mean tax share than 35% to be SME, actually the same classification as Ruedaand Pontusson (2000) is reached.20

Estimating separate models for LME and SME (see Table 6), we find that market forces havea much stronger effect on inequality in LME. A change in taxes as share of GDP has onlya negative (albeit not significant) effect on inequality in LMEs, where this level tends to begenerally low. The lack of significance of the tax share variable in the subsamples can thus beexplained by the fact that it is correlated with the dummy variable defining LME and SME. Thesignal content of the within-group variability in tax shares in each one of the two groups is thusnot powerful enough as to lead to significant parameter estimates for this variable. Similarly,it is only in LMEs that an increase of non-oil-producing LDC imports is associated with an

19The empirical result that rising minimum wages diminish wage inequality is documented in Koeniger et al.(2007).

20Rueda and Pontusson (2000) and Esping-Andersen (1990) include less countries in their studies than wedo. Rueda and Pontusson (2000) label the following countries as LME: Australia, United States, Canada,Switzerland, Japan and United Kingdom. The following group forms the SME subsample: Germany, Finland,Austria, Belgium, Netherlands, Denmark, Norway, Sweden. Italy and France are labelled as mixed economies.

15

increase of income inequality. This suggests that SMEs are capable of mitigating the effect oftrade on income inequality more efficiently than LMEs.21

6.4 Inequality and the Power of Trade Unions

Our results concerning an overall income-equalizing effect of trade unions confirm the findingsof Mahler (2004), Koeniger et al. (2007) and Beramendi and Cusack (2008). Mahler (2004)finds that higher levels of wage coordination lower inequality and more so when measured asdisposable income inequality. Disposable income inequality is additionally lowered by higherlevels of union density. Beramendi and Cusack (2008), Koeniger et al. (2007) and Rueda andPontusson (2000) confirm empirically the influence of unions on inequality. Pontusson et al.(2002) analyze the dynamics of wage inequality in the upper and lower halves of the wagedistribution separately and find that higher levels of labour organization through unions isespecially raising the relative wages of the low-paid unskilled workers in the lower half of thedistribution.


Following Baccaro (2011), we take a step further and study potential changes in the effect ofunionization on income distribution over time. In the context of shrinking numbers of unionizedworkers and of rising globalization, it has been argued that internationalization renders labourdemand curves more elastic, which increasingly lowers the bargaining power of unions. Weestimate a model in which time dummies – for each one of the periods in our sample – areinteracted with the union density variable. Our results, presented in column 1 of Table 7,indicate that in parallel to the loss of bargaining power throughout the 40 year period of thestudy, a rise in union density is less significantly linked to a decline of inequality. In column2 of the same table the results of a model that analyzes the impact of non-oil-exporting LDCimports over time are shown. In contrast to the effect of unions, the effect of trade increasesover time, becoming significant in the early nineties and the strongest in the last years of thesample.

7 Conclusions

We present a fully-fledged empirical analysis of the determinants of income inequality in thedeveloped world, using the widest-ranging panel dataset available and a multitude of poten-tial determinants implied by theory. As such, this study takes a comprehensive approach tounveiling the drivers of inequality in the developed world and thus improves on the existingliterature, which tends to concentrate on individual drivers.

The main finding of our empirical analysis is that low-wage imports to developed countries tendto worsen income inequality. This finding is very robust and the variation in the trade variableexplains a large fraction of the within-country variation of income inequality. As opposed

21This result is not driven by differences in the overall openness of the countries in the two subsamples. Thevariability of the trade variable is comparable across subgroups.

16

to most of the other contributions to the literature, our analysis concentrates on a variablewhich is constructed in such a way as to mimic the type of trade flows which are impliedby the theoretical setting of the Heckscher-Ohlin model. The Stolper-Samuelson theorem as acorollary of the Heckscher-Ohlin model predicts this result, thus our study confirms the standardprediction of trade theory.

In addition, we find that government size, the interaction of technology and education, de-mocratization processes and the reshaping of the labour market by unions have played animportant role in explaining the dynamics of income inequality in the developed world over thelast decades. In addition, the effect of government size appears to be particularly relevant forliberal market economies. While the size of the effect of unionization has decreased over time,income inequality tends to react more strongly to international trade in the recent past.

17

References

Acemoglu, D. (2002). Technical change, inequality, and the labor market. Journal of EconomicLiterature 40 (1), 7–72.

Acemoglu, D. (2003). Patterns of skill premia. Review of Economic Studies 70 (2), 199–230.

Anderson, T. W. and C. Hsiao (1981). Estimation of dynamic models with error components.Journal of the American Statistical Association 76 (375), 598–606.

Anderson, T. W. and C. Hsiao (1982). Formulation and estimation of dynamic models usingpanel data. Journal of Econometrics 18 (1), 47–82.

Arellano, M. and S. Bond (1991). Some tests of specification for panel data: Monte carloevidence and an application to employment equations. Review of Economic Studies 58 (2),277–297.

Armingeon, K., S. Engler, P. Potolidis, M. Gerber, and P. Leimgruber (2010). Comparativepolitical data set 1960-2008. Institute of Political Science, University of Berne.

Armingeon, K. and C. Romana (2007). Comparative data set for 28 post-communist countries,1989-2007. Institute of Political Science, University of Berne.

Atkinson, A. B. and A. Brandolini (2001). Promise and Pitfalls in the Use of ’Secondary’Data-Sets: Income Inequality in OECD Countries as a Case Study. Journal of EconomicLiterature 39 (3), 771–799.

Baccaro, L. (2011). Labor, globalization and inequality: Are trade unions still redistributive?Research in the Sociology of Work 22, 213–285.

Beramendi, P. and T. R. Cusack (2008). Diverse disparities: The politics and economics of wage,market, and disposable income inequalities. Political Research Quarterly 62 (2), 257–275.

Blau, F. D. and L. M. Kahn (1996). International differences in male wage inequality: Institu-tions versus market forces. Journal of Political Economy 104 (4), 791–837.

Blundell, R. and S. Bond (1998). Initial conditions and moment restrictions in dynamic paneldata models. Journal of Econometrics 87 (1), 115–143.

Bussmann, M., I. De Soysa, and J. R. Oneal (2005). The effect of globalization on nationalincome inequality. Comparative Sociology 3, 285–312.

Calderon, C., A. Chong, and R. O. Valdes (2005). Labor market regulations and incomeinequality: Evidence for a panel of countries. Central Banking, Analysis, and EconomicPolicies Book Series , 221–279.

Checchi, D. and C. Garcia-Penalosa (2008). Labour market institutions and income inequality.Economic Policy 23, 601–649.

Checchi, D. and C. Garcia-Penalosa (2010). Labour Market Institutions and the PersonalDistribution of Income in the OECD. Economica 77 (307), 413–450.

18

Deininger, K. and L. Squire (1996). A new data set measuring income inequality. World BankEconomic Review 10 (3), 565–91.

Dreher, A. and N. Gaston (2008). Has globalization increased inequality? Review of Interna-tional Economics 16 (3), 516–536.

Esping-Andersen, G. (1990). The Three Worlds of Welfare Capitalism. Princeton UniversityPress.

Feenstra, R. C. (1998). Integration of trade and disintegration of production in the globaleconomy. Journal of Economic Perspectives 12 (4), 31–50.

Figini, P. and H. Gorg (2011). Does foreign direct investment affect wage inequality? anempirical investigation. World Economy 34 (9), 1455–1475.

Freeman, R. (1982). Union wage practices and wage dispersion within establishments. Industrialand Labor Relations Review 36 (1), 3–21.

Galbraith, J. K., B. Halbach, A. Malinowska, A. Shams, and W. Zhang (2014). The UTIPGlobal Inequality Data Sets 1963-2008. Updates, Revisions and Quality Checks. UTIP Work-ing Paper 68 .

Galbraith, J. K. and H. Kum (2005). Estimating the inequality of household incomes: Astatistical approach to the creation of a dense and consistent global data set. Review ofIncome and Wealth 51 (1), 115–143.

Hall, P. A. and D. Soskice (2001). Varieties of Capitalism: The Institutional Foundations ofComparative Advantage. Oxford University Press.

Helpman, E. and P. Krugman (1987). Market Structure and Foreign Trade: Increasing Returns,Imperfect Competition, and the International Economy. The MIT Press.

Heston, A., R. Summers, and B. Aten (2011). Penn world table version 7.0. Center for Inter-national Comparisons of Production, Income and Prices at the University of Pennsylvania.

Im, K. S., M. H. Pesaran, and Y. Shin (2003). Testing for unit roots in heterogeneous panels.Journal of Econometrics 115 (1), 53–74.

Isham, J., M. Woolcock, L. Pritchett, and G. Busby (2005). The varieties of resource experience:Natural resource export structures and the political economy of economic growth. World BankEconomic Review 19 (2), 141–174.

Jaumotte, F., S. Lall, and C. Papageorgiou (2008). Rising income inequality: Technology, ortrade and financial globalization? International Monetary Fund Working Papers 08 (185).

Koeniger, W., M. Leonardi, and L. Nunziata (2007). Labour market institutions and wageinequality. Industrial and Labor Relations Review 60 (3), 340–356.

Krugman, P. (2000). Technology, trade and factor prices. Journal of International Eco-nomics 50 (1), 51–71.

19

Krugman, P., B. P. Bosworth, and R. N. Cooper (1999). Domestic policies in a global economy.Brookings Trade Forum, 73–103.

Kuncic, A. (2014). Institutional quality dataset. Journal of Institutional Economics 10 (01),135–161.

Kuznets, S. S. (1955). Economic growth and income inequality. American Economic Review 45,1–28.

Lee, W. and G. A. Slotsve (2001). Globalization, world democratization, and the dynamics ofincome distribution across nations. Northern Illinois University.

Lundberg, M. and L. Squire (2003). The simultaneous evolution of growth and inequality.Economic Journal 113 (487), 326–344.

Mahler, V. A. (2004). Economic globalization, domestic politics, and income inequality in thedeveloped countries: A cross-national study. Comparative Political Studies 37 (9), 1025–1053.

Meschi, E. and M. Vivarelli (2009). Trade and income inequality in developing countries. WorldDevelopment 37 (2), 287–302.

Muller, E. N. (1995). Economic determinants of democracy. American Sociological Re-view 60 (6), 966–982.

Oliver, R. (2008). Diverging developments in wage inequality: Which institutions matter?Comparative Political Studies 41, 1551–1582.

Pinkovskiy, M. L. (2013). World welfare is rising: Estimation using nonparametric bounds onwelfare measures. Journal of Public Economics 97, 176–195.

Pontusson, J., D. Rueda, and C. R. Way (2002). Comparative political economy of wagedistribution: The role of partisanship and labour market institutions. British Journal ofPolitical Science 32 (02), 281–308.

Reuveny, R. and Q. Li (2003). Economic openness, democracy, and income inequality. Com-parative Political Studies 36 (5), 575–601.

Rueda, D. and J. Pontusson (2000). Wage inequality and varieties of capitalism. World Poli-tics 52 (3), 350–383.

Sala-i Martin, X. (2006). The World Distribution of Income: Falling Poverty and ... Conver-gence, Period. Quarterly Journal of Economics 121 (2), 351–397.

Samanni, M., J. Teorell, S. Kumlin, and B. Rothstein (2010). The QoG social policydataset, version 11nov10. University of Gothenburg: The Quality of Government Institute -http://www.qog.pol.gu.se.

Samuelson, P. A. (1949). International factor-price equalisation once again. Economic Jour-nal 59 (234), 181–197.

Western, B. and J. Rosenfeld (2011). Unions, norms, and the rise in U.S. wage inequality.American Sociological Review 76 (4), 513–537.

20

Figure 1: Average Gini index (EHII dataset) for 32 developed economies. Unweighted averageversus population-weighted average.

Figure 2: Gini index (EHII dataset) for 32 developed economies. Countries ascending by medianinequality.

21

02

46

810

Impo

rts fr

om D

evel

opin

g C

ount

ries

(no

natu

ral r

esou

rces

) in

% o

f GD

P

MexicoTu

rkeySpa

in

Canad

a

Greece

Portug

alIsr

ael

New Zea

land

Icelan

dIta

ly

France

United

States

Austra

lia

Norway

United

Kingdo

mIre

land

Poland

Sweden

South

Korea

Luxe

mbourg

Austria

FinlandJa

panChile

Belgium

Hunga

ry

Denmark

German

y

Slovak

Repub

lic

Netherl

ands

Czech

Repub

lic

Sloven

ia

Figure 3: Imports from LDCs (without oil and raw material exporters) as % of GDP.

Figure 4: Mean Gini index for different groups of countries

22

Table 1: Baseline model with different trade variables(1) (2) (3) (4) (5) (6) (7)

EHII lagged 0.599*** 0.619*** 0.609*** 0.621*** 0.611*** 0.609*** 0.673***

(21.67) (22.71) (22.10) (22.65) (22.11) (22.05) (25.61)

GDP per capita -2.180 -4.977 -2.134 -5.000 -2.673 -2.246 -3.182

(-0.62) (-1.43) (-0.60) (-1.43) (-0.75) (-0.63) (-0.80)

GDP per capita2 0.0558 0.210 0.0518 0.210 0.0801 0.0570 0.159

(0.31) (1.17) (0.28) (1.17) (0.43) (0.31) (0.76)

GDP growth -0.00974 0.00222 -0.00207 0.00303 -0.000671 -0.00157 -0.00315

(-0.65) (0.15) (-0.14) (0.20) (-0.05) (-0.11) (-0.20)

Tax share of GDP -0.0607*** -0.0607*** -0.0704*** -0.0582*** -0.0664*** -0.0692*** -0.0284

(-4.01) (-3.92) (-4.37) (-3.73) (-4.15) (-4.30) (-1.49)

Education -1.105** -0.771* -0.831* -0.713* -0.788* -0.823* -0.528

(-2.52) (-1.78) (-1.93) (-1.65) (-1.83) (-1.91) (-0.87)

Imports from LDCs, no oil 0.457***

(3.90)

Imports from LDCs 0.0760*

(1.94)

Total Imports 0.0276***

(2.85)

Exports to LDCs 0.0487

(1.07)

Total Exports 0.0215**

(2.27)

Trade Openness 0.0131*** 0.0121**

(2.64) (2.08)

Constant 33.19** 44.57*** 32.43* 44.63*** 36.93** 35.13** 29.87

(1.99) (2.67) (1.91) (2.66) (2.18) (2.07) (1.58)

Observations 849 849 849 849 849 849 849

Countries 32 32 32 32 32 32 32

Period Effects Yes Yes Yes Yes Yes Yes Yes

Sargan χ2 591.0 591.0 588.4 591.2 591.5 590.3 OLS

p-value Sargan 0.0893 0.0891 0.1017 0.0882 0.0867 0.0926 -

p-value AR (1) 0.0003 0.0002 0.0002 0.0002 0.0002 0.0002 -

p-value AR (2) 0.0149 0.0129 0.0134 0.0131 0.0133 0.0135 -

(*), [**], {***} indicate significance at the (10 %), [5 %], {1 %} level respectively. Estimation by System-GMM . Dependent variable

is the Gini index from Estimated Household Income Inequality (EHII). z-values in brackets. AR (1) and AR (2) refer to the p-value

test statistic for first- and second-order autocorrelation on the disturbances of the first-differences equations. p-value Sargan is the

p-value of the test statistic for the validity of the overidentifying restriction. Decade dummies included in all models.

23

Table 2: Alternative specifications of the baseline model(1) (2) (3) (4) (5) (6) (7)

EHII lagged 0.622*** 0.577*** 0.442*** 0.599*** 0.628*** 0.562*** 0.564***

[18.71] [16.22] [10.80] [21.67] [18.89] [20.96] [17.29]

GDP per capita -2.180 -4.868 -5.605* -6.515

[-0.62] [-0.95] [-1.69] [-0.86]

GDP per capita2 0.0558 0.184 0.253 0.327

[0.31] [0.69] [1.45] [0.82]

GDP growth -0.00974 0.000946 -0.00693 0.00144

[-0.65] [0.04] [-0.43] [0.07]

Education -1.105** -1.309 -0.478 -1.514**

[-2.52] [-1.45] [-0.90] [-2.23]

Tax share of GDP -0.169*** -0.0607*** -0.0242 -0.0439**

[-7.54] [-4.01] [-1.62] [-2.49]

Public Expenditure -0.0334***

Share of GDP [-4.08]

Imports from LDCs, 0.408*** 0.753*** 0.457*** 0.221* 0.330*** 0.238*

no oil [3.11] [5.19] [3.90] [1.68] [2.85] [1.72]

Union density -0.0309*** -0.0270***

[-5.13] [-3.71]

Inflation 0.0282*** 0.0246***

[4.74] [3.73]

FDI Stock Inward 0.0157*

[1.70]

FDI Stock Outward -0.0148

[-0.88]

Constant 14.42*** 15.33*** 25.22*** 33.19** 46.96* 49.39*** 53.99

[11.57] [11.62] [12.85] [1.99] [1.89] [3.07] [1.47]

Observations 1075 1031 850 849 564 744 447

Countries 32 32 32 32 32 30 29

Period Effects Yes Yes Yes Yes Yes Yes Yes

Sargan χ2 226.3 231.4 190.0 591.0 378 725.3 530.1

p-value Sargan 0.0000 0.0000 0.0000 0.0893 0.0142 0.1162 0.0529

p-value AR (1) 0.0005 0.0006 0.0002 0.0003 0.0029 0.0004 0.0007

p-value AR (2) 0.0650 0.0369 0.0140 0.0149 0.1296 0.0453 0.0852





24

Table 3: Alternative specifications with further control variables1 2 3 4

EHII lagged 0.564*** 0.561*** 0.563*** 0.561***

[20.92] [20.92] [20.92] [20.81]

GDP per capita -4.684 -4.656 -5.383 -5.433

[-1.24] [-1.30] [-1.60] [-1.63]

GDP per capita2 0.208 0.201 0.239 0.231

[1.07] [1.07] [1.35] [1.28]

GDP growth -0.00479 -0.00548 -0.00712 -0.00975

[-0.30] [-0.34] [-0.44] [-0.60]

Education -0.466 -0.505 -0.489 -0.485

[-0.87] [-0.94] [-0.92] [-0.91]

Tax share of GDP -0.0235 -0.0191 -0.0233 -0.0240

[-1.56] [-1.16] [-1.54] [-1.61]

Imports from LDCs, no oil 0.348*** 0.334*** 0.332*** 0.345***

[2.84] [2.88] [2.86] [2.90]

Union density -0.0320*** -0.0312*** -0.0307*** -0.0304***

[-5.14] [-5.17] [-5.07] [-4.87]

Inflation 0.0285*** 0.0276*** 0.0279*** 0.0278***

[4.79] [4.62] [4.68] [4.68]

Dependency ratio 0.00838

[0.59]

Population growth 0.0962

[0.72]

Immigration 0.0668

[0.36]

Solow Residual 0.363

[0.33]

Constant 44.11** 44.90*** 48.46*** 45.47**

[2.35] [2.60] [2.97] [2.33]

Observations 744 744 744 744

Countries 30 30 30 30

Period Effects Yes Yes Yes Yes

Sargan χ2 732.7 724.4 724.1 726.4

p-value Sargan 0.0829 0.1209 0.1225 0.1109

p-value AR (1) 0.0004 0.0004 0.0004 0.0004

p-value AR (2) 0.0450 0.0451 0.0459 0.0458





25

Table 4: Models with interaction of TFP and education(1) (2)

EHII lagged 0.561*** 0.560***

(20.81) (20.82)

GDP per capita -5.433 1.425

(-1.63) (0.31)

GDP per capita2 0.231 -0.129

(1.28) (-0.53)

GDP growth -0.00975 -0.00776

(-0.60) (-0.48)

Union density -0.0304*** -0.0316***

(-4.87) (-5.03)

Inflation 0.0278*** 0.0271***

(4.68) (4.57)

Tax share of GDP -0.0240 -0.0161

(-1.61) (-1.05)

Imports from LDCs, no oil 0.345*** 0.340***

(2.90) (2.85)

Education -0.485 -49.22**

(-0.91) (-2.25)

Solow Residual 0.363 -8.554**

(0.33) (-2.06)

Solow Residual × Education 4.114**

(2.23)

Constant 45.47** 118.2***

(2.33) (3.10)

Observations 744 744

Countries 30 30

Period Effects Yes Yes

Sargan χ2 726.4 722.1

p-value Sargan 0.1109 0.1336

p-value AR (1) 0.0004 0.0004

p-value AR (2) 0.0458 0.0466





26

Table 5: Models including political regime change(1) (2)

EHII lagged 0.410*** 0.459***

(11.04) (12.59)

GDP per capita -8.146 -5.235

(-1.38) (-0.84)

GDP per capita2 0.378 0.190

(1.17) (0.57)

GDP growth -0.00217 -0.0159

(-0.13) (-0.92)

Democracy -0.167***

(-3.95)

Democracy × Ex-Soviet dummy 3.795*

(1.95)

Democracy × Western Europe dummy -0.232***

(-4.74)

Democracy × Rest of the World dummy -0.0546

(-0.80)

Tax share of GDP 0.0560* 0.0597*

(1.74) (1.84)

Education 2.414** 1.690*

(2.52) (1.73)

Imports from LDCs, no oil 0.259 0.380**

(1.38) (2.04)

Constant 59.11** 46.84*

(2.19) (1.65)

Observations 765 765

Countries 29 29

Period Effects Yes Yes

Sargan χ2 521.1 511.9

p-value Sargan 0.0317 0.3012

p-value AR (1) 0.0005 0.0005

p-value AR (2) 0.0185 0.0191

(*), [**], {***} indicate significance at the (10 %), [5 %], {1 %} level respectively. Estimation by Difference-GMM . Dependent

variable is the Gini index from Estimated Household Income Inequality (EHII). z-values in brackets. AR (1) and AR (2) refer to

the p-value test statistic for first- and second-order autocorrelation on the disturbances of the first-differences equations. p-value

Sargan is the p-value of the test statistic for the validity of the overidentifying restriction. Decade dummies included in all models.

27

Table 6: Income inequality determinants: Liberal versus social market economies(1) (2) (3)

Sample Total SME LME

EHII lagged 0.599*** 0.714*** 0.648***

[21.67] [21.22] [20.41]

GDP per capita -2.180 -13.16 0.974

[-0.62] [-1.25] [0.25]

GDP per capita2 0.0558 0.626 -0.100

[0.31] [1.17] [-0.49]

GDP growth -0.00974 0.00961 -0.00494

[-0.65] [0.37] [-0.29]

Tax share of GDP -0.0607*** 0.00293 -0.0325

[-4.01] [0.11] [-1.27]

Education -1.105** 0.0113 -1.272***

[-2.52] [0.02] [-2.75]

Non-oil LDC imports 0.457*** 0.150 0.302**

[3.90] [1.16] [2.00]

Constant 33.19** 78.44 17.51

[1.99] [1.52] [0.94]

Observations 849 326 523

Countries 32 14 18

Period Effects Yes Yes Yes

Sargan χ2 591.0 416.5 485.6

p-value Sargan 0.0893 0.5256 0.8272

p-value AR (1) 0.0003 0.0037 0.0085

p-value AR (2) 0.0149 0.0233 0.3056

SME are the Social Market Economies - defined as those countries with a mean tax share above 35% of the country’s GDP. LME

are the Liberal Market Economies and are all remaining economies. (*), [**], {***} indicate significance at the (10 %), [5 %], {1 %}level respectively. Estimation by System-GMM . Dependent variable is the Gini index from Estimated Household Income Inequality

(EHII). z-values in brackets. AR (1) and AR (2) refer to the p-value test statistic for first- and second-order autocorrelation on the

disturbances of the first-differences equations. p-value Sargan is the p-value of the test statistic for the validity of the overidentifying

restriction. Decade dummies included in all models.

28

Table 7: Determinants of income inequality: Time-varying parameters for trade and uniondensity

(1) (2)

EHII lagged 0.541*** EHII lagged 0.534***

[19.96] [19.90]

GDP per Capita -1.258 GDP per Capita -2.927

[-0.35] [-0.88]

GDP per capita2 0.0249 GDP per capita2

0.0972

[0.13] [0.55]

GDP growth 0.000661 GDP growth 0.00375

[0.04] [0.23]

Tax Share of GDP -0.0374** Tax Share of GDP -0.0188

[-2.49] [-1.26]

Education -0.724 Education -0.789

[-1.36] [-1.49]

Non-oil LDC Imports 0.232* Union Density -0.0303***

[1.80] [-5.08]

Union Density Non-oil LDC Imports

interacted with time dummy: interacted with time dummy:

1963 -1966 -0.0590*** 1963 -1966 -1.314***

[-6.50] [-3.34]

1967 -1971 -0.0497*** 1967 -1971 -0.909**

[-5.65] [-2.42]

1971 - 1974 -0.0429*** 1971 - 1974 -0.605**

[-5.87] [-2.30]

1975 - 1978 -0.0388*** 1975 - 1978 -0.385*

[-5.60] [-1.75]

1979 - 1982 -0.0328*** 1979 - 1982 -0.142

[-4.97] [-0.73]

1983 - 1986 -0.0309*** 1983 - 1986 -0.366

[-4.67] [-1.60]

1987 - 1990 -0.0261*** 1987 - 1990 0.0403

[-3.87] [0.23]

1991 - 1994 -0.0228*** 1991 - 1994 0.373**

[-3.48] [2.40]

1995 - 1998 -0.0207*** 1995 - 1998 0.391***

[-2.79] [3.03]

1999 - 2002 -0.0130 1999 - 2002 0.467***

[-1.35] [3.51]

30.36* Constant 39.59**

[1.76] [2.46]

Observations 744 Observations 744

Countries 30 Countries 30

Period Effects Yes Period Effects Yes

Sargan χ2 724.1 Sargan χ2 718.5

p-value Sargan 0.1069 p-value Sargan 0.1305

p-value AR (1) 0.0005 p-value AR (1) 0.0004

p-value AR (2) 0.0612 p-value AR (2) 0.0806





29

Appendix

Data Interpolation in the EHII Dataset

The lack of consistent panel datasets for countrywide income inequality measures has led severalresearchers to provide reconstructed inequality data based on regression analysis. The EHIIdataset (Galbraith and Kum, 2005) is one of the most widely used examples of such databases.Galbraith and Kum (2005) utilize the information contained in the UTIP-UNIDO measureof dispersion of manufacturing pay in the framework of regression analysis in order to obtainestimates of gross household income inequality (as measured by the Gini index).

Galbraith and Kum (2005) start by estimating a regression model linking the measure of in-equality in Deininger and Squire (1996) (DSit) to the manufacturing pay dispersion variable(MDit), a set of dummy variables which account for the characteristics of the observation inthe Deininger and Squire (1996) dataset (Dj,it) and other socioeconomic variables (Xk,it). Themodel used is thus given by

logDSit = α + δMDit +∑j

ρjDj,it +∑k

θkXk,it + εit

where εit is the regression error, assumed to fulfil the necessary conditions for the standardlinear regression model.

Galbraith and Kum (2005) obtain parameter estimates for this regression model using a set ofdummy variables that identify the data source as Dj,it variables (whether the inequality measureis based on income or expenditure, whether it is measured at the individual or household leveland whether it is a based on a gross or net variable) and the share of employment in themanufacturing sector to total population. With the aid of these parameter estimates, themissing data in DSit are interpolated as the corresponding conditional expectation, utilizingthe available information for the independent variables of the model. A similar regression-based approach to the imputation of missing data for income inequality can be found in Sala-iMartin (2006). Among other, the recent contribution by Pinkovskiy (2013), which attemptsat measuring the change of world inequality and welfare over time, reports that the results ofthe analysis are similar when using survey data from the World Income Inequality Database(WIID) or the Gini index provided by the EHII dataset.

30

Data Sources

• The Deninger-Squire income inequality dataset is available athttp://go.worldbank.org/UVPO9KSJJ0.

• The University of Texas Inequality Project publishes the data on income inequality athttp://utip.gov.utexas.edu/data.html.

• The Correlates of War data is available at: http://correlatesofwar.org.

• The FDI dataset is available (for observations starting in 1980) at: http://unctadstat.unctad.org/TableViewer/tableView.aspx?ReportId=89.

• The Database on Institutional Characteristics of Trade Unions, Wage Setting, State In-tervention and Social Pacts, maintained by Jelle Visser at the Amsterdam Institute forAdvanced Labour Studies (AIAS) is freely available at: http://www.uva-aias.net/208.This information was complemented for our analysis with data from OECD EmploymentDatabase. The data is available online: http://www.oecd.org/document/34/0,3343,

en264933927409171541111,00.html.

• The Polity IV dataset is available and documented online at: http://www.systemicpeace.org/polity/polity4.htm.

• Data on the tax share of GDP are available at the OECD Tax Database at http://www.oecd.org/document/60/0,3746,en26493453319424601111,00.html.

• The data on the share of public expenditures on GDP is taken from the Statistics ofPublic Expenditure for Economic Development (SPEED) database, compiled by the In-ternational Food Policy Research Institute (IFPRI) online at http://www.ifpri.org/

book-39/ourwork/programs/priorities-public-investment/speed-database.

• Unemployment data from Armingeon et al. (2010) and Armingeon and Romana (2007) areavailable online at http://www.ipw.unibe.ch/content/team/klausarmingeon, whilefemale labour participation data are sourced from the OECD’s Labour Force Statistics(MEI), available online at: http://stats.oecd.org/.

• The Barro-Lee education data are available online at: www.barrolee.com.

• Total factor productivity data from the conference board available at https://www.

conference-board.org/data/economydatabase/ were used for the robustness analysis.

31

Inequality Measures and Tax Shares: Descriptive Statistics

Table A1: Descriptive statistics, Gini index (EHII) for our sample of 32 developed countriesCountry Observations Mean Std.Dev. Min Max

Australia 39 33.78 0.255 30.66 38.83

Austria 37 35.03 0.114 33.31 40.42

Belgium 35 36.06 0.133 34.06 38.18

Canada 39 36.07 0.123 34.26 38.19

Chile 38 44.29 0.259 38.89 48.58

Czech Republic 33 23.84 0.320 19.81 33.16

Denmark 36 30.95 0.483 29.76 31.98

Finland 38 31.72 0.159 27.95 33.66

France 21 35.01 0.819 33.92 36.42

Germany 38 32.77 0.126 29.65 36.48

Greece 36 40.41 0.204 33.64 42.87

Hungary 38 31.09 0.482 23.84 40.02

Iceland 29 36.14 0.277 31.34 40.95

Ireland 38 38.26 0.278 33.59 46.23

Israel 39 39.87 0.284 35.31 43.34

Italy 34 35.73 0.244 31.73 40.09

Japan 39 37.44 0.229 35.07 43.12

Luxembourg 38 32.78 0.158 29.77 35.17

Mexico 31 41.08 0.185 35.86 44.03

Netherlands 38 32.82 0.153 29.65 35.10

New Zealand 36 35.21 0.262 32.19 42.14

Norway 39 32.30 0.145 29.29 35.72

Poland 31 30.50 0.404 25.78 38.51

Portugal 32 39.93 0.200 35.13 43.54

Slovak Republic 7 33.93 0.298 30.19 36.70

Slovenia 16 31.30 0.346 24.95 35.32

South Korea 39 39.21 0.222 36.07 43.36

Spain 40 38.48 0.295 30.97 41.45

Sweden 38 27.98 0.111 25.89 30.21

Turkey 38 43.23 0.237 40.19 47.97

United Kingdom 34 30.35 0.363 26.45 36.77

United States 38 37.01 0.116 34.79 38.40

32

Table A2: Descriptive statistics, level and change in Gini index (EHII) since the eightiesMean between Last observation Change in percent since Last year of

1980 and 1986 of EHII data 1980-1986 EHII observation

Iceland 39,62 34,27 -13,51 1996

Belgium 36,89 36,70 -0,53 1997

Chile 45,88 46,51 1,36 2000

Denmark 31,04 31,51 1,50 1998

United States 37,58 38,31 1,92 2001

Israel 41,39 42,90 3,64 2001

Portugal 38,37 39,95 4,09 2000

Spain 35,05 36,62 4,48 2002

Greece 39,72 41,56 4,62 1998

Canada 36,23 38,00 4,89 2001

Luxembourg 32,59 34,40 5,54 2000

France 34,39 36,42 5,88 2000

Norway 31,65 34,16 7,92 2001

Sweden 26,62 28,96 8,78 2000

Finland 30,24 32,92 8,84 2000

Italy 33,33 36,31 8,93 2000

Mexico 39,95 43,83 9,70 2000

Netherlands 31,68 35,10 10,78 2000

Germany 32,80 36,48 11,20 2000

Turkey 42,10 47,97 13,93 2000

Australia 32,97 38,35 16,29 2001

Austria 34,74 40,42 16,34 2000

New Zealand 33,34 38,94 16,79 1998

Japan 36,50 43,12 18,12 2001

Ireland 37,23 44,80 20,33 2000

United Kingdom 29,28 36,77 25,58 2000

Slovenia 24,95* 34,42 37,96 2002

Poland 26,91 38,51 43,08 2000

Hungary 27,79 39,87 43,45 2000

Czech Republic 22,10 33,16 50,01 1995

* As Slovenia was not an independent country in the beginning of the 1980s the value of Slovenia is from 1987 - which is the first

available from the EHII dataset.

1020

3040

50

Tota

l Tax

as

shar

e of

GD

P

Turke

y

Mexico

South

KoreaChile

Greece

Portug

alJa

pan

United

States

Austra

liaSpa

inIre

land

Icelan

d

Slovak

Repub

lic

New Zea

land

Canad

a

Poland

United

Kingdo

mItaly

German

yIsr

ael

Czech

Repub

lic

Luxe

mbourg

Sloven

ia

Hunga

ry

Austria

Netherl

ands

Finland

Norway

France

Belgium

Denmark

Sweden

Figure A1: Total taxes as share of GDP

33

Documents

ePubWU Institutional RepositoryMax Roser and Jesus Crespo Cuaresma Why is Income Inequality Increasing in the Developed World? Article (Accepted for Publication) (Refereed) Original